HPI/my/core/hpi_compat.py
karlicoss d3f9a8e8b6
core: migrate code to benefit from 3.9 stuff (#401)
for now keeping ruff on 3.8 target version, need to sort out modules as well
2024-10-19 20:55:09 +01:00

260 lines
7.9 KiB
Python

"""
Contains various backwards compatibility/deprecation helpers relevant to HPI itself.
(as opposed to .compat module which implements compatibility between python versions)
"""
from __future__ import annotations
import inspect
import os
import re
from collections.abc import Iterator, Sequence
from types import ModuleType
from typing import TypeVar
from . import warnings
def handle_legacy_import(
parent_module_name: str,
legacy_submodule_name: str,
parent_module_path: list[str],
) -> bool:
###
# this is to trick mypy into treating this as a proper namespace package
# should only be used for backwards compatibility on packages that are convernted into namespace & all.py pattern
# - https://www.python.org/dev/peps/pep-0382/#namespace-packages-today
# - https://github.com/karlicoss/hpi_namespace_experiment
# - discussion here https://memex.zulipchat.com/#narrow/stream/279601-hpi/topic/extending.20HPI/near/269946944
from pkgutil import extend_path
parent_module_path[:] = extend_path(parent_module_path, parent_module_name)
# 'this' source tree ends up first in the pythonpath when we extend_path()
# so we need to move 'this' source tree towards the end to make sure we prioritize overlays
parent_module_path[:] = parent_module_path[1:] + parent_module_path[:1]
###
# allow stuff like 'import my.module.submodule' and such
imported_as_parent = False
# allow stuff like 'from my.module import submodule'
importing_submodule = False
# some hacky traceback to inspect the current stack
# to see if the user is using the old style of importing
for f in inspect.stack():
# seems that when a submodule is imported, at some point it'll call some internal import machinery
# with 'parent' set to the parent module
# if parent module is imported first (i.e. in case of deprecated usage), it won't be the case
args = inspect.getargvalues(f.frame)
if args.locals.get('parent') == parent_module_name:
imported_as_parent = True
# this we can only detect from the code I guess
line = '\n'.join(f.code_context or [])
if re.match(rf'from\s+{parent_module_name}\s+import\s+{legacy_submodule_name}', line):
importing_submodule = True
# click sets '_HPI_COMPLETE' env var when it's doing autocompletion
# otherwise, the warning will be printed every time you try to tab complete
autocompleting_module_cli = "_HPI_COMPLETE" in os.environ
is_legacy_import = not (imported_as_parent or importing_submodule)
if is_legacy_import and not autocompleting_module_cli:
warnings.high(
f'''\
importing {parent_module_name} is DEPRECATED! \
Instead, import from {parent_module_name}.{legacy_submodule_name} or {parent_module_name}.all \
See https://github.com/karlicoss/HPI/blob/master/doc/MODULE_DESIGN.org#allpy for more info.
'''
)
return is_legacy_import
def pre_pip_dal_handler(
name: str,
e: ModuleNotFoundError,
cfg,
requires: Sequence[str] = (),
) -> ModuleType:
'''
https://github.com/karlicoss/HPI/issues/79
'''
if e.name != name:
# the module itself was imported, so the problem is with some dependencies
raise e
try:
dal = _get_dal(cfg, name)
warnings.high(
f'''
Specifying modules' dependencies in the config or in my/config/repos is deprecated!
Please install {' '.join(requires)} as PIP packages (see the corresponding README instructions).
'''.strip(),
stacklevel=2,
)
except ModuleNotFoundError:
dal = None
if dal is None:
# probably means there was nothing in the old config in the first place
# so we should raise the original exception
raise e
return dal
def _get_dal(cfg, module_name: str):
mpath = getattr(cfg, module_name, None)
if mpath is not None:
from .utils.imports import import_dir
return import_dir(mpath, '.dal')
else:
from importlib import import_module
return import_module(f'my.config.repos.{module_name}.dal')
V = TypeVar('V')
# named to be kinda consistent with more_itertools, e.g. more_itertools.always_iterable
class always_supports_sequence(Iterator[V]):
"""
Helper to make migration from Sequence/List to Iterable/Iterator type backwards compatible in runtime
"""
def __init__(self, it: Iterator[V]) -> None:
self._it = it
self._list: list[V] | None = None
self._lit: Iterator[V] | None = None
def __iter__(self) -> Iterator[V]: # noqa: PYI034
if self._list is not None:
self._lit = iter(self._list)
return self
def __next__(self) -> V:
if self._list is not None:
assert self._lit is not None
delegate = self._lit
else:
delegate = self._it
return next(delegate)
def __getattr__(self, name):
return getattr(self._it, name)
@property
def _aslist(self) -> list[V]:
if self._list is None:
qualname = getattr(self._it, '__qualname__', '<no qualname>') # defensive just in case
warnings.medium(f'Using {qualname} as list is deprecated. Migrate to iterative processing or call list() explicitly.')
self._list = list(self._it)
# this is necessary for list constructor to work correctly
# since it's __iter__ first, then tries to compute length and then starts iterating...
self._lit = iter(self._list)
return self._list
def __len__(self) -> int:
return len(self._aslist)
def __getitem__(self, i: int) -> V:
return self._aslist[i]
def test_always_supports_sequence_list_constructor() -> None:
exhausted = 0
def it() -> Iterator[str]:
nonlocal exhausted
yield from ['a', 'b', 'c']
exhausted += 1
sit = always_supports_sequence(it())
# list constructor is a bit special... it's trying to compute length if it's available to optimize memory allocation
# so, what's happening in this case is
# - sit.__iter__ is called
# - sit.__len__ is called
# - sit.__next__ is called
res = list(sit)
assert res == ['a', 'b', 'c']
assert exhausted == 1
res = list(sit)
assert res == ['a', 'b', 'c']
assert exhausted == 1 # this will iterate over 'cached' list now, so original generator is only exhausted once
def test_always_supports_sequence_indexing() -> None:
exhausted = 0
def it() -> Iterator[str]:
nonlocal exhausted
yield from ['a', 'b', 'c']
exhausted += 1
sit = always_supports_sequence(it())
assert len(sit) == 3
assert exhausted == 1
assert sit[2] == 'c'
assert sit[1] == 'b'
assert sit[0] == 'a'
assert exhausted == 1
# a few tests to make sure list-like operations are working..
assert list(sit) == ['a', 'b', 'c']
assert [x for x in sit] == ['a', 'b', 'c'] # noqa: C416
assert list(sit) == ['a', 'b', 'c']
assert [x for x in sit] == ['a', 'b', 'c'] # noqa: C416
assert exhausted == 1
def test_always_supports_sequence_next() -> None:
exhausted = 0
def it() -> Iterator[str]:
nonlocal exhausted
yield from ['a', 'b', 'c']
exhausted += 1
sit = always_supports_sequence(it())
x = next(sit)
assert x == 'a'
assert exhausted == 0
x = next(sit)
assert x == 'b'
assert exhausted == 0
def test_always_supports_sequence_iter() -> None:
exhausted = 0
def it() -> Iterator[str]:
nonlocal exhausted
yield from ['a', 'b', 'c']
exhausted += 1
sit = always_supports_sequence(it())
for x in sit:
assert x == 'a'
break
x = next(sit)
assert x == 'b'
assert exhausted == 0
x = next(sit)
assert x == 'c'
assert exhausted == 0
for _ in sit:
raise RuntimeError # shouldn't trigger, just exhaust the iterator
assert exhausted == 1